EXPERT SYSTEMS

EXPERT SYSTEMS
Expert systems are knowledge-based systems that provide expertise, similar to that of
experts in a restricted application area. (Kasabov, 1998)
• Whenever we need help in everyday life, we consult an expert.
• An expert is a person who has profound knowledge of the subject.
• This knowledge is a result of the expert’s capabilities and experience.
• The expert looks at the problems, makes some inferences, and presents the results.
• As an end customer, we follow the results and assume that the results are true.
• The expert also continues to update knowledge and to learn from past cases.
• Experts provide end-to-end solutions to meet our needs.
• Expert system (ES) design is motivated by this concept of experts.
• In this type of design, we try to make systems that can act autonomously.
• Once these systems are presented with the conditions as inputs, they work over the solutions.
• The solution given by these systems is then implemented.
• These systems are a complete design of whatever it might take to draw conclusion once the problem is given.
• To reach the conclusions may require preprocessing, processing algorithms, or any other procedures.
• These systems may require a historical database or a knowledge base to learn from
Architecture of Expert System
 Architecture of Expert System
➢ User interface: This is where the user interacts with the system. The user can select options, give
inputs, and receive outputs. The user interface is designed to keep all the functionality available
in the system in some or the other way.
➢ Knowledge base: This is the pool of knowledge available in the system. Knowledge may be
represented in many ways. Standard AI and soft-computing techniques are used for this purpose.
This pool of knowledge can be easily queried to solve common problems per the requirements of
the user.
➢ Database: This module is a collection of data. All past or historical data are stored here. The
database itself has the potential to generate a huge amount of knowledge once any learning
technique, or any other means, is applied over it. Analysis or learning of these data generates the
knowledge used by the knowledge base.
➢ Inference engine: This module is responsible for interpreting the knowledge or its inference. It
assembles and presents the output to the user in the form desired. This engine controls the
system’s overall working.
➢ Knowledge acquisition: This module traps data or inputs and adds them to the pool so the system
can learn new things from the data that have been presented. Knowledge acquisition is
responsible for making the system robust against the newer inputs. This increases the system’s
performance.
➢ Explanation: This module is responsible for tracing the execution of the entire process. It records
how the various inferences were made and why the specific rules were fired. In expert systems, it
is responsible for the reasoning behind the generation of the output from the inputs.
 DESIGN OF EXPERT SYSTEM (ES)
➢ The basic objective and motive of the ES must be very clear.
➢ The ES must be able to replace the existing system.
➢ The first major task is identification of inputs. Inputs should be :
well known.
relevant to the system.
➢ The ES should be clear about the way in which these inputs are to be used by the system.
➢ There should be suitable interface for the inputs to be taken from the user.
➢ Similarly the outputs must be clearly known.
➢ The number and kind of outputs that the system is expected to produce must be known.
➢ The outputs generated must be of interest to the user or of relevance to the system.
➢ There also must be a good interface to present the output to the user.
➢ All the data needed for a process must be made available so that the system can perform and the
functionality can be met.
➢ We further need to work out the correct method for storing and processing knowledge, the types of AI and
soft-computing tools that need to be used.
➢ The ES must be able to perform in the given constraints and in the best possible way. It also must be cross
verified. Once the implementation is done, it may be checked for performance and errors
TYPES OF PROBLEMS

➢ Classification

➢ Functional Approximation

➢ Optimizations
 CLASSIFICATION
➢ In classification problems, inputs are mapped to a class of outputs.
➢ The output set consists of discrete values or classes.
➢ The system is supposed to see the inputs and select one of the classes to which the input belongs.
Every input can map only to a single class.
➢ Hence, once an input is presented, if the system maps it to the correct class, the answer is
regarded as correct. If the system maps it to the incorrect class, the answer is incorrect.
➢ In the case of classification problems, the system tries to learn the differences between the
various classes based on the inputs presented.
➢ The system’s performance may be measured as the percentage of inputs correctly classified.
➢ It can be observed that the system’s performance depends on the number of classes.
➢ Consider the example of speech recognition. Suppose we are making an automatic phone dialer
that dials the number we speak. We know that the person speaking will say any one of the ten
digits from 0 to 9. Hence, the output for every input can be any one of the ten digits. The system
is supposed to identify the digit. Here, we can say that there are ten classes, and the classification
problem is to map the input to any of the ten classes.
 FUNCTIONAL APPROXIMATION
➢ In functional approximation problems, the output is continuous and is an unknown function of the inputs.
➢ For every input, there is a specific output in the entire output space.
➢ The system is supposed to predict the unknown function that maps the inputs to the outputs.
➢ The system tries to predict the same unknown function in the learning phase.
➢ Whenever the input is provided, the system calculates the outputs.
➢ In these systems, it is very difficult for any input to get the precisely correct output.
➢ But the output given by the system is usually very close to the actual output.
➢ The closeness depends on the input given and the kind of learning the system was able to perform.
➢ The system’s performance may be measured by the deviation of the actual output to the desired output.
➢ The purpose of such a system’s design and learning is to make it possible to imitate the function that maps
the inputs to the outputs.
➢ Here performance may be measured by the mean percent deviation or root mean square error.
➢ Consider this example: We are trying to make a system that controls a boiler’s temperature. The system
takes as its input the present temperature and the quantity of fluid in the boiler. The system then outputs
the fuel to be added per second to make the boiler work. The more fuel is added, the higher the heat is
applied and the higher the temperature. This system is a functional approximation system that tries to map
the inputs and the outputs by an otherwise unknown function
OPTIMIZATIONS
➢In optimization, we are given a function, known as the objective
function.
➢The aim is to minimize or maximize the value of the objective
function by adjusting its various parameters.
➢Each combination of parameters marks a solution that may be good
or bad, depending on the value of the objective function.
➢Soft-computing techniques generate the best possible parameter set
that gives the best values of the objective function considering the
time constraints.